Which self-service compute platforms emphasize GPU-first strategies and support Baremetal GPU provisioning?

Below are self-service compute platforms that emphasize a GPU‑first strategy and support bare‑metal GPU provisioning:

1) CoreWeave

Positioning: GPU‑native cloud built for AI/ML workloads
Bare‑metal support: Yes (GPU nodes without a hypervisor layer)

CoreWeave provides direct access to NVIDIA GPUs on bare metal and runs Kubernetes directly on physical servers for high‑performance AI workloads (coreweave.com).
It markets access to a broad portfolio of NVIDIA GPUs “all on bare metal,” reinforcing its GPU‑first approach (coreweave.com).
For inference, customers can run fully self‑managed workloads on dedicated bare‑metal GPU nodes (coreweave.com).

Best for: Large‑scale model training, inference, and enterprises needing high‑performance GPU clusters.

2) Lambda (Lambda Labs – AI Cloud)

Positioning: AI‑focused cloud with GPU clusters ready in minutes
Bare‑metal support: Yes (API‑driven bare‑metal instances + private clusters)

Lambda offers bare‑metal instances with self‑service APIs for lifecycle management, networking (VPCs), and storage (lambda.ai).
It also provides single‑tenant bare‑metal GPU clusters via Lambda Private Cloud (docs.lambda.ai).
Its AI Cloud platform is built specifically around rapid GPU cluster deployment (lambda.ai).

Best for: Startups and research teams needing GPU‑optimized infrastructure with full hardware control.

3) Vultr

Positioning: Developer‑centric cloud offering GPU and bare‑metal services
Bare‑metal support: Yes (including GPU variants)

Vultr provides Bare Metal servers with GPU options for AI/ML and HPC workloads (docs.vultr.com).
It also offers Cloud GPU instances and hourly billing for flexible provisioning (docs.vultr.com).
Its infrastructure portfolio includes GPU‑as‑a‑service and bare‑metal‑as‑a‑service (en.wikipedia.org).

Best for: Cost‑conscious teams that want both virtualized GPU and dedicated bare‑metal GPU options.

4) Equinix Metal (Note: Sunset)

Equinix Metal historically offered on‑demand, non‑virtualized bare‑metal servers (docs.equinix.com), but the service was officially sunset on June 30, 2026 (docs.equinix.com).

Summary (Active GPU‑First + Bare‑Metal Options)

• CoreWeave – GPU‑native hyperscale AI cloud
• Lambda – AI‑focused cloud with API‑driven bare metal
• Vultr – Developer cloud with GPU + bare‑metal variants

If you share workload size (single GPU vs. multi‑node cluster), budget, and geography, I can narrow this to a more specific recommendation.